Obstruction-Aware Bluetooth Low Energy Indoor Positioning

On the High Fluctuations of Received Signal Strength Measurements With BLE Signals for Contact Tracing and Proximity Detection

“…Also, the NLOS distributions have higher or equal spread than LOS ones in most of the cases, even though the NLOS data sets contained less recordings than the LOS ones. This result contradicts observations in [44], where NLOS obstructions caused by walls were identified when the standard deviation of RSS measurements in a window was lower than a fixed threshold. Although UPB and TAU data sets were acquired using the same model of Raspberry Pis, measurements acquired at TAU had a smaller spread than those from UPB even in LOS, which points once more to the instability of RSS measurements.…”

“…Therefore, obstructions between the transmitter and the receiver will typically cause fluctuations in the RSS. There are several research works [27], [43], [44] that investigated the effect of shadowing on the BLE RSS with applications in proximity detection or localization. [27] proposed artificial neural network (ANN) models for detecting human-body shadowing and compensating RSS values to improve distance measurements or localization based on the BLE RSS.…”

“…In [44], the authors proposed a NLOS detection method based on the variance of the BLE RSS. The algorithm is able to detect when a concrete wall is blocking the direct path between the transmitter and the receiver with an accuracy of 76.25 % based on a fixed threshold of the RSS variance, below which the signal is classified as being acquired in NLOS.…”

“…Indeed, other literature results have shown that the path-loss parameters used in different works vary widely. For instance, although a path-loss exponent between 2.4-2.6 is frequently recommended [27], in [44] the path-loss exponent was set to 1.8 for LOS scenarios and 2.2 for NLOS ones. In the survey part of [2], the path-loss exponents extracted from various research papers varied between 0.63-2.32 and TABLE II suggests that in some cases (NLOS, UPB) this value might be even lower.…”

“…In proximity-detection or RSS-based localization applications, obstructions will most of the time lead to inaccurate distance or location estimates. Therefore, multiple solutions have been proposed to correct RSS-based ranges by detecting the NLOS condition [27], [43], [44] with the caveat that such solutions might not generalize easily, as our measurements show, or that large data sets might be necessary to extract features that improve classification. In contact-tracing applications, such instabilities can lead to false alarms or failures in detecting potentially unsafe interactions.…”